Fault positioning method based on large model, intelligent agent, device and equipment

Abstract

The invention provides a fault positioning method based on a large model, an intelligent agent, a device and equipment, and relates to the technical field of computers, in particular to the technical fields of cloud computing, deep learning, large models and the like. According to the specific implementation scheme, fault description information is obtained; performing information extraction on the fault description information by using a pre-trained fault positioning agent to obtain key information, and determining troubleshooting contents and an execution sequence of the troubleshooting contents based on the fault description information; and calling a target detection tool corresponding to each troubleshooting content by using a pre-trained fault positioning agent, carrying out troubleshooting based on the key information according to the execution sequence to obtain a troubleshooting result corresponding to each troubleshooting content, and obtaining a fault positioning result based on each troubleshooting result, and the fault positioning efficiency is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, specifically to technical fields such as cloud computing, deep learning, and large models, and in particular to a fault location method, intelligent agent, device, and equipment based on a large model. Background Art

[0002] CDN (Content Delivery Network) is a distributed network consisting of edge node servers distributed in different regions, which is built and covered on the existing bearer network. CDN enables users to obtain the required content nearby, solves the problem of Internet network congestion, and improves the response speed of users' website access. With the rapid development of the number of user services and the Internet era, the number and frequency of user access failures are increasing. If the failure is not solved in time, it is easy to affect the quality of CDN service. Summary of the invention

[0003] The present disclosure provides a fault location method, intelligent agent, device and equipment based on a large model.

[0004] According to one aspect of the present disclosure, a fault location method based on a large model is provided, comprising:

[0005] Get fault description information;

[0006] Using a pre-trained fault location agent, extract information from the fault description information to obtain key information;

[0007] Using a pre-trained fault location agent, determine the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information;

[0008] Using the pre-trained fault location agent, calling the target detection tool corresponding to each of the troubleshooting contents, performing troubleshooting based on the key information in the execution order, and obtaining troubleshooting results corresponding to each of the troubleshooting contents;

[0009] A pre-trained fault location agent is used to obtain a fault location result based on each of the troubleshooting results.

[0010] According to another aspect of the present disclosure, a method for training a fault localization agent is provided, wherein the fault localization agent includes a first large language model, a second large language model, and a third large language model, and the method includes:

[0011] Acquire sample fault description information, annotated key information corresponding to the sample fault description information, and a first sample prompt word;

[0012] Based on the sample fault description information, the first sample prompt word and the labeled key information, a first large language model is trained; the first large language model is used to extract information from the sample fault description information under the guidance of the first sample prompt word;

[0013] Obtaining the annotated fault troubleshooting content corresponding to the sample fault description information and the annotated execution order of each annotated fault troubleshooting content, as well as a second sample prompt word;

[0014] Based on the sample fault description information, the second sample prompt word, the annotated fault troubleshooting content and the annotated execution order of each annotated fault troubleshooting content, a second language model is trained; the second language model is used to determine the fault troubleshooting content corresponding to the sample fault description information and its execution order under the guidance of the second sample prompt word; the annotated fault troubleshooting content includes information about the detection tool required to be used for the annotated fault troubleshooting content;

[0015] Invoke the sample detection tools corresponding to the troubleshooting contents of the annotations respectively according to the annotation execution order, perform troubleshooting based on the annotation key information, and obtain the sample troubleshooting results corresponding to the troubleshooting contents of the annotations;

[0016] Obtaining the labeled fault location result corresponding to each of the sample troubleshooting results and the third sample prompt word;

[0017] Based on the sample troubleshooting results, the third sample prompt words and the labeled fault location results, a third language model is trained; the third language model is used to summarize the fault problems in the sample troubleshooting results under the guidance of the third sample prompt words.

[0018] According to another aspect of the present disclosure, a large model-based fault location device is provided, comprising:

[0019] An information acquisition module is used to obtain fault description information;

[0020] An information extraction module, used to extract information from the fault description information using a pre-trained fault location agent to obtain key information;

[0021] A content determination module, used to determine the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information by using a pre-trained fault location agent;

[0022] A first troubleshooting module is used to use a pre-trained fault location intelligent agent to call a target detection tool corresponding to each of the troubleshooting contents, perform troubleshooting based on the key information in the execution order, and obtain troubleshooting results corresponding to each of the troubleshooting contents;

[0023] The fault location module is used to obtain a fault location result based on each of the above-mentioned troubleshooting results by using a pre-trained fault location agent.

[0024] According to another aspect of the present disclosure, a training device for a fault localization agent is provided, wherein the fault localization agent includes a first large language model, a second large language model, and a third large language model, and the training device includes:

[0025] A first acquisition module, used to acquire sample fault description information, annotated key information corresponding to the sample fault description information, and a first sample prompt word;

[0026] A first training module, used for training a first large language model based on the sample fault description information, the first sample prompt word and the annotation key information; the first large language model is used for extracting information from the sample fault description information under the guidance of the first sample prompt word;

[0027] A second acquisition module is used to acquire the annotated fault troubleshooting content corresponding to the sample fault description information and the annotated execution order of each annotated fault troubleshooting content, as well as a second sample prompt word;

[0028] A second training module is used to train a second large language model based on the sample fault description information, the second sample prompt word, the labeled fault troubleshooting content and the labeled execution order of each labeled fault troubleshooting content; the second large language model is used to determine the fault troubleshooting content corresponding to the sample fault description information and its execution order under the guidance of the second sample prompt word; the labeled fault troubleshooting content contains information about the detection tool required to use the labeled fault troubleshooting content;

[0029] A second troubleshooting module is used to call the sample detection tools corresponding to each of the annotated troubleshooting contents respectively according to the annotated execution order, perform troubleshooting based on the annotated key information, and obtain sample troubleshooting results corresponding to each of the annotated troubleshooting contents;

[0030] A third acquisition module is used to acquire the marked fault location result corresponding to each of the sample troubleshooting results and the third sample prompt word;

[0031] The third training module is used to train a third language model based on the sample troubleshooting results, the third sample prompt words and the labeled fault location results; the third language model is used to summarize the fault problems of each sample troubleshooting result under the guidance of the third sample prompt words.

[0032] According to another aspect of the present disclosure, there is provided an intelligent agent, comprising:

[0033] An input module, used for receiving fault description information;

[0034] A processing module, configured to extract information from the fault description information received by the input module to obtain key information; determine the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information; call the target detection tool corresponding to each fault troubleshooting content, perform fault troubleshooting based on the key information in the execution order, and obtain troubleshooting results corresponding to each fault troubleshooting content; and obtain a fault location result based on each troubleshooting result;

[0035] An output module is used to output the fault location result obtained by the processing module.

[0036] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0037] at least one processor; and

[0038] a memory communicatively connected to the at least one processor; wherein,

[0039] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform any of the above methods.

[0040] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute any one of the above methods.

[0041] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein the computer program implements any of the above methods when executed by a processor.

[0042] The disclosed embodiments improve the efficiency of fault location.

[0043] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0045] Figure 1 It is a flow chart of a fault location method based on a large model according to the present disclosure;

[0046] Figure 2 is a schematic diagram of fault location according to the fault location intelligent agent disclosed in the present invention;

[0047] Figure 3 is a schematic diagram of a training process of a fault location agent according to the present disclosure;

[0048] Figure 4 It is a flow chart of a training method of a fault location intelligent agent based on a large model according to the present disclosure;

[0049] Figure 5 is a structural schematic diagram of a fault location device based on a large model according to the present disclosure;

[0050] Figure 6 is a structural schematic diagram of a training device for a fault location intelligent agent according to the present disclosure;

[0051] Figure 7 is a structural schematic diagram of an intelligent agent according to the present disclosure;

[0052] Figure 8 It is a block diagram of an electronic device used to implement the large model-based fault location method or the training method of the fault location agent disclosed in the embodiment of the present disclosure. DETAILED DESCRIPTION

[0053] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0054] First, the relevant terms or concepts in the embodiments of the present disclosure are explained:

[0055] AI Agent: It is an intelligent agent system with a large language model as the core controller. Its essence is an agent system that controls the large language model to solve user problems. It relies on the large language model as its core decision-making and processing unit, and can use the logical reasoning ability of the large language model to solve user problems. In practical applications, the agent can be a software or hardware entity that can act autonomously.

[0056] LLM (Large Language Model) refers to a deep learning model trained on massive text data, which has strong expressive and generalization capabilities. It can not only generate natural language text, but also deeply understand the meaning of the text and handle various natural language tasks such as text summarization, question answering, translation, etc.

[0057] In the related art, the method used for fault location in CDN scenarios is: manual location. Specifically, after the user reports the fault problem, the back-end operation and maintenance personnel will troubleshoot the problem based on their experience, such as the log data of user requests, the quality of CDN nodes that provide services to users, etc., and use network detection tools such as ping (Packet Internet Groper, Internet packet detector), mtr (my traceroute, network link routing test tool), traceroute (route tracking) to detect network quality, etc., to troubleshoot the problem and locate the fault.

[0058] The manual positioning method is inefficient due to the different experience of different operation and maintenance personnel, and the troubleshooting time is usually inversely proportional to the experience, which makes the troubleshooting efficiency low and easily leads to user complaints.

[0059] In order to improve the efficiency of fault location, the embodiments of the present disclosure provide a large model-based fault location method, a training method, a device and equipment for a fault location intelligent agent.

[0060] The fault location method based on the large model provided by the embodiment of the present disclosure is as follows: Figure 1 As shown, the following steps are included:

[0061] S101, obtaining fault description information;

[0062] S102, using a pre-trained fault location agent to extract information from the fault description information to obtain key information;

[0063] S103, using a pre-trained fault location agent to determine the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information;

[0064] Among them, step S102 and step S103 can be executed sequentially or in parallel.

[0065] S104, using the pre-trained fault location intelligent agent, calling the target detection tool corresponding to each fault troubleshooting content, performing fault troubleshooting based on key information in the execution order, and obtaining the troubleshooting results corresponding to each fault troubleshooting content;

[0066] S105, using the pre-trained fault location agent to obtain a fault location result based on each troubleshooting result.

[0067] In the disclosed embodiment, for the fault description information fed back by the user, a pre-trained fault location agent is directly used to extract information from the fault description information to obtain key information, and the fault troubleshooting content and the execution order of each fault troubleshooting content are determined based on the fault description information. Then, the target detection tool corresponding to each fault troubleshooting content is called, and fault troubleshooting is performed based on the key information in the execution order to obtain the troubleshooting results corresponding to each fault troubleshooting content. Then, the fault location result is obtained based on each troubleshooting result, that is, the troubleshooting process of the operation and maintenance personnel is integrated into the function of the fault location agent, and the fault is located by the fault location agent, thereby improving the efficiency of fault location.

[0068] The large model-based fault location method and the training method of the fault location agent provided in the embodiments of the present disclosure are described in detail below.

[0069] The fault location method based on a large model provided in the embodiment of the present disclosure can be applied to fault location in CDN scenarios, and of course can also be applied to other network fault location scenarios. The embodiment of the present disclosure only takes the CDN scenario as an example for illustration. Among them, the fault location method based on a large model can be implemented by a fault location agent, which can be a hardware entity, such as an intelligent robot, or a software application deployed in an electronic device, such as a client device or a server device.

[0070] Figure 1 The fault location method based on the large model is implemented by a fault location agent. The embodiment of the present disclosure pre-trains the fault location agent. Specifically, the large language model is trained according to the fault reporting data in the daily CDN scenario, and then the trained large language model is used to construct the fault location agent. The fault location agent directly calls the trained large language model during use. In one example, in order to improve the training efficiency of the model and save costs, the large language model trained by the embodiment of the present disclosure can be a fine-tuning training of the pre-trained large language model.

[0071] In the disclosed embodiment, the fault location agent realizes fault location through four stages. One stage is the information extraction stage, which specifically extracts information from the fault description information input to the fault location agent to obtain key information in the fault description information; one stage is the planning stage of the troubleshooting process and sequence, which determines the troubleshooting content and the execution order of each troubleshooting content based on the fault description information input therein; another stage is the tool execution stage, that is, calling the target detection tool corresponding to each troubleshooting content, and troubleshooting based on the key information in the execution order to obtain the troubleshooting results corresponding to each troubleshooting content; and another stage is the conclusion summary stage, which summarizes the problems based on the troubleshooting results obtained by the troubleshooting to obtain the fault location results. Specifically, the training method of the fault location agent is described in detail below.

[0072] Figure 1 In the fault location method based on the large model shown, the fault description information can be a device alarm or a fault information reported by a customer received by the operation and maintenance department. In an example, the fault description information can include but is not limited to: the time when the abnormality occurred, the abnormality description information, the abnormality state, the location where the abnormality occurred, the abnormal domain name, etc.

[0073] The following embodiments of the present disclosure are described by taking fault location in a CDN scenario as an example.

[0074] For example, the fault description information may be: A customer's domain name xxx.com reported a fault, and the connection quality in region B deteriorated.

[0075] The user or operation and maintenance personnel inputs the fault description information into the pre-trained fault location agent, which receives the fault description information and then extracts the information from the fault description information to obtain key information; determines the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information; calls the target detection tool corresponding to each fault troubleshooting content, performs troubleshooting based on the key information in the execution order, and obtains the troubleshooting results corresponding to each fault troubleshooting content; obtains the fault location result based on each troubleshooting result. For the fault location agent, the fault description information is the query statement of the user or operation and maintenance personnel.

[0076] The fault location agent extracts the domain name, region, etc. contained in the fault description information to obtain key information.

[0077] In a possible implementation, the above step S102 uses a pre-trained fault location agent to extract information from the fault description information to obtain key information, including:

[0078] The pre-trained fault location agent is used to call the first language model, and the fault description information is input into the first language model to extract information under the guidance of the first prompt word to obtain key information.

[0079] The fault location agent is constructed by the trained large language model, and the trained large language model is directly called in the process of fault location. For information extraction, the first large language model is trained in advance according to the sample fault description information, the first sample prompt word and the labeled key information corresponding to the sample fault description information. Accordingly, the trained first large language model is used to extract information from the sample fault description information under the guidance of the first sample prompt word.

[0080] When the fault location agent receives the fault description information input by the user or the operation and maintenance personnel, it directly calls the trained first language model and inputs the fault description information into the first language model. The first language model extracts information under the guidance of the first prompt word to obtain key information. The first prompt word is used to instruct the first language model to extract key information from the fault description information.

[0081] The first prompt word may be pre-set or obtained when training the first language model. The key information indicates key information that can be used for troubleshooting, including but not limited to: time, node name, domain name, operator, region, etc.

[0082] Exemplarily, the first prompt word may be:

[0083] You are a key information extraction assistant who can extract key information from the query statement and concatenate it into a string. The following information is known:

[0084] Please help me extract the following keywords based on the question. It can be one or more keywords, or it can be empty. Multiple keywords can be separated by ","

[0085] 1. Extract the node name, the key is node, and the specifications are as follows: [Node specification example]

[0086] 2. Extract the domain name, the key is domain, and the specifications are as follows: [Domain name specification example]

[0087] 3. Extract the operator, the key is isp, the operator does not contain the corresponding province name, such as China Mobile, China Unicom, China Telecom, etc.

[0088] 4. Extract provinces or municipalities, key is prov

[0089] Construct data based on the results. The output can only be in the following JSON format. There should be no comments in the JSON result:

[0090] {{

[0091] "node": "Node name",

[0092] "domain": "Domain name",

[0093] "isp": "operator",

[0094] "prov": "province"

[0095] }}

[0096] Make sure the result can be parsed by Python's json.loads

[0097] This is the user's question: {goal}.

[0098] The fault location agent calls the first language model, and the first language model extracts key information from the fault description information under the guidance of the first prompt word to obtain the above-mentioned key information such as node name, domain name, operator, region, etc.

[0099] In the disclosed embodiment, a pre-trained fault location agent is used to extract key information from the fault description information through the trained first language model combined with the first prompt word, thereby improving the accuracy of key information extraction so that the fault location can be more accurately performed later.

[0100] In a possible implementation, the above step S103 uses a pre-trained fault location agent to determine the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information, including:

[0101] The pre-trained fault location agent is used to call the second largest language model, and the fault description information is input into the second largest language model to determine the fault troubleshooting content and the execution order of the fault troubleshooting content under the guidance of the second prompt word.

[0102] In the embodiment of the present disclosure, for planning the troubleshooting process and sequence, the second largest language model is trained in advance according to the sample fault description information, the second sample prompt word, the labeled troubleshooting content corresponding to the sample fault description information, and the labeled execution sequence of each labeled troubleshooting content. Accordingly, the trained second largest language model is used to determine the troubleshooting content corresponding to the sample fault description information and its execution sequence under the guidance of the second sample prompt word.

[0103] When the fault location agent receives the fault description information input by the user or the operation and maintenance personnel, it calls the trained second language model and inputs the fault description information into the second language model. The second language model determines the fault troubleshooting content and the execution order of the fault troubleshooting content under the guidance of the second prompt word. The second prompt word is used to instruct the second language model to plan the fault troubleshooting process and sequence for the fault description information, so as to determine the fault troubleshooting content corresponding to the fault description information and the execution order of the fault troubleshooting content.

[0104] In one example, the second prompt word can be determined based on historical troubleshooting experience. The second prompt word can be pre-set, or it can be obtained by training the second largest language model. For example, the troubleshooting process included in the historical troubleshooting experience includes: determining the CDN node of the service, querying the service request log during the fault reporting period, querying whether the basic indicators of the service domain name are normal, querying whether there are problems with the quality of the CDN node of the service, and testing whether there are network problems with the CDN service node through ping, mtr, traceroute, etc. Correspondingly, the second prompt word can be set based on the troubleshooting process to prompt the second largest language model to think about the troubleshooting content and the execution order of each troubleshooting content.

[0105] Exemplarily, the second prompt word may be:

[0106] Your task is to come up with a short plan that helps me locate the user's problem in a few steps, using at most one of the provided tools for each step. You can create new steps with the help of the following tools.

[0107] Executing a tool can be thought of as: using the tool to test the target, where the target is a sub-target of the original target. You will get a summary of the tool's output.

[0108] You can use {tool_specification}

[0109] {current_date_and_time}

[0110] Construct data according to the planning results. The output can only be in the following JSON format. There should be no comments in the JSON result:

[0111] {{

[0112] "steps": [

[0113] "Step1: xxxxx",

[0114] "Step2: xxxxxx"

[0115] ],

[0116] "order": "Execution order logic format"

[0117] }}

[0118] Make sure the result can be parsed by Python's json.loads

[0119] This is the problem reported by the user: {goal}.

[0120] The second prompt word includes information about tools that can be called, and the tools can be specifically listed in {tool_specification}. Correspondingly, the troubleshooting content includes information about the detection tools required for the troubleshooting content.

[0121] Exemplarily, the troubleshooting content determined by the second largest language model may include:

[0122] Step 1: Use the indicator detection tool to detect the bandwidth, number of requests, and status code indicator information of the domain name xxx.com in region B;

[0123] Step 2: Use the scheduling detection tool to detect the current coverage changes of the domain name xxx.com in region B and the current bearer information;

[0124] Step 3: Use the indicator detection tool to detect the current detection success rate, detection speed, and node CPU pressure indicator information of the xxx.com coverage node;

[0125] Step 4: Use the change control tool to detect recent abnormal changes in the nodes covered by xxx.com;

[0126] Step 5: Use the log search tool to retrieve the log information of the domain name xxx.com at the above node.

[0127] The execution order of each troubleshooting content can be: (Step1 AND Step2) -> (Step3 AND Step4AND Step5). This execution order means: Step1 and Step2 are executed in parallel, Step3, Step4 and Step5 are executed in parallel, and (Step1 AND Step2) and (Step3 AND Step4 AND Step5) are executed in series.

[0128] Among them, the information of the detection tools required for troubleshooting content includes: indicator detection tools, scheduling detection tools, change management tools, log retrieval tools, etc.

[0129] In the disclosed embodiment, a pre-trained fault location agent is utilized to determine the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information through the trained second largest language model combined with the second prompt word, thereby improving the accuracy of the fault troubleshooting process and sequence planning. Based on this, fault troubleshooting can be performed accurately, and then fault location can be performed more accurately.

[0130] In a possible implementation, the determined troubleshooting contents have levels and priorities, and troubleshooting contents at the same level are executed in parallel, while troubleshooting contents at different priorities are executed serially.

[0131] The level and priority of each troubleshooting content can be known through the determined troubleshooting content and its execution order. Exemplarily, the execution order of the above troubleshooting contents is: (Step1 AND Step2) -> (Step3 ANDStep4 AND Step5), and accordingly, the troubleshooting contents Step1 and Step2 have the same level, and Step3, Step4 and Step5 have the same level, and Step1 and Step2 have a higher priority than Step3, Step4 and Step5.

[0132] In the disclosed embodiment, the level and priority of each troubleshooting content can be known through the determined troubleshooting content and its execution order, so as to more clearly determine the troubleshooting status of each troubleshooting link in the troubleshooting process.

[0133] In a possible implementation, the above step S104 uses a pre-trained fault location agent to call the target detection tool corresponding to each troubleshooting content, performs troubleshooting based on key information in the execution order, and obtains troubleshooting results corresponding to each troubleshooting content, including:

[0134] Using the pre-trained fault location agent, determine the target detection tool corresponding to the troubleshooting content according to the information of the detection tool contained in the troubleshooting content;

[0135] The target detection tools corresponding to each troubleshooting content are called respectively in the execution order, and troubleshooting is performed based on the key information to obtain the troubleshooting results corresponding to each troubleshooting content.

[0136] When extracting key information from the fault description information and determining the troubleshooting content and the execution order of each troubleshooting content, since the troubleshooting content contains information about the detection tools required for the troubleshooting content, the target detection tool corresponding to the troubleshooting content can be determined directly based on the information about the detection tools contained in the troubleshooting content. Further, the target detection tools corresponding to each troubleshooting content are called in the execution order, and troubleshooting is performed in combination with the key information to obtain the troubleshooting results corresponding to each troubleshooting content.

[0137] For example, as in the above troubleshooting content Step 1: Use the indicator detection tool to detect the bandwidth, number of requests, and status code indicator information of the domain name xxx.com currently connected to China Unicom in region B. According to the information of the detection tool: indicator detection tool, it is determined that the target detection tool corresponding to Step 1 is the indicator detection tool, and its corresponding execution order is at the front, so the indicator detection tool is first called to perform troubleshooting in combination with the key information domain name (xxx.com), region (region B), and operator (China Unicom), and the corresponding troubleshooting results are obtained, such as the bandwidth, number of requests, and status code indicator information of the domain name xxx.com currently connected to China Unicom in region B.

[0138] In one example, when calling the target detection tool corresponding to each troubleshooting content and performing troubleshooting in combination with key information, a calling instruction for the target detection tool can be generated first, and then the target detection tool can be called through the calling interface of the target detection tool to perform troubleshooting.

[0139] In the disclosed embodiment, the fault location intelligent agent can directly call the target detection tool corresponding to the fault troubleshooting content to perform fault troubleshooting, thereby achieving efficient fault troubleshooting.

[0140] In a possible implementation, the target detection tool corresponding to the troubleshooting content is called, and troubleshooting is performed based on the key information to obtain troubleshooting results corresponding to the troubleshooting content, including:

[0141] Call the target detection tool corresponding to the troubleshooting content, so that the target detection tool performs troubleshooting based on the key information to obtain the detection result data, and call the fourth language model to analyze the detection result data to obtain the troubleshooting result corresponding to the troubleshooting content.

[0142] In one example, the internal execution logic of the detection tool can call the large language model to analyze the troubleshooting result data obtained by troubleshooting to obtain the troubleshooting result corresponding to the troubleshooting.

[0143] Exemplarily, taking the target detection tool as an indicator detection tool, the indicator detection tool can detect all queries and analyses involving basic indicators. In actual applications, indicators usually include [bandwidth], [qps (queries-per-second, query rate per second)], [request success rate], [status code], etc. In one case, the fault location agent calls the indicator detection tool, and the indicator detection tool directly performs troubleshooting based on the key information determined above. In another case, the fault location agent calls the indicator detection tool, and the indicator detection tool cannot accurately determine the detection object based on the key information determined above. At this time, the key information determined above is vectorized, and then the similarity is calculated with the vector representation of the description information of the detection tool stored in the preset database. The detection tool corresponding to the similarity greater than the set threshold (such as 0.8, 0.9 or 0.95, etc.) is determined as the final detection tool, and the determined final detection tool is called to perform troubleshooting to obtain the detection result data.

[0144] Furthermore, the fourth language model is called to analyze the detection result data to obtain the troubleshooting results corresponding to the troubleshooting content. The fourth language model can be trained based on the sample detection result data and its corresponding annotated troubleshooting results. The indicator is generally a time series curve. During the execution of the indicator detection tool, the fourth language model can be called to analyze the detection result data, and the obtained detection result data can be displayed to the user in the form of a curve, and the curve can be analyzed for sudden increase, sudden decrease, discrete point analysis, data variance analysis, etc., and finally the troubleshooting results, that is, the indicator conclusion, are returned to the fault location intelligent agent.

[0145] Exemplarily, taking the target detection tool as a log retrieval tool as an example, the log retrieval tool is used to perform log retrieval based on the customer's fault reporting time and customer domain name, and extract key columns based on the retrieved logs. These columns may include [client], [request time], [status code], [sent data size], [request processing time], [HTTP (Hypertext Transfer Protocol) request header], [TCP (transmission control protocol) indicators (such as RTT (round-trip time), CWND (congestion window))], etc., and the log data is analyzed based on the extracted information to see if there is a certain pattern, such as abnormal data concentrated on a certain client, User-Agent concentrated on a certain URL (uniform resource locator; uniform resource locator), etc.

[0146] In the CDN scenario, if the user request does not hit the cache, a back-to-source request will be made. For example, when the log retrieval tool requests a specified URL, it will perform an overall request chain analysis. If it does not hit the cache, it will continue to query the second-level and third-level nodes until it hits the cache or returns to the customer's origin station. At the same time, the request chain analysis will be performed when summarizing the results, such as the reason for the back-to-source failure and whether the back-to-source request chain meets expectations.

[0147] Correspondingly, the fault location agent calls the log retrieval tool, which performs log retrieval based on the time and domain name contained in the key information to obtain the retrieval data, and then calls the fourth language model to analyze the retrieval data to obtain the corresponding troubleshooting results.

[0148] In the disclosed embodiment, the fault location intelligent agent calls the target detection tool corresponding to the troubleshooting content to perform troubleshooting. During the troubleshooting process, the target detection tool can call the fourth language model to analyze the detection result data, and obtain the troubleshooting results corresponding to the troubleshooting content, thereby realizing intelligent analysis and summary of the detection result data.

[0149] In a possible implementation, the above step S105 uses a pre-trained fault location agent to obtain a fault location result based on each troubleshooting result, including:

[0150] The pre-trained fault location agent is used to call the third language model, and the troubleshooting results corresponding to each fault troubleshooting content are input into the third language model. The fault problem is summarized under the guidance of the third prompt word to obtain the fault location result.

[0151] In the embodiment of the present disclosure, for summarizing the conclusions, a third language model is pre-trained based on the sample screening results, the third sample prompt words, and the labeled fault location results corresponding to the sample screening results. Accordingly, the trained third language model is used to summarize the fault problems of each sample screening result under the guidance of the third sample prompt words.

[0152] The fault location agent extracts key information from the fault description information, determines the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information, and calls the target detection tool corresponding to each fault troubleshooting content. After performing troubleshooting based on the key information in the execution order and obtaining the troubleshooting results corresponding to each fault troubleshooting content, the third language model is further called to input the troubleshooting results corresponding to each fault troubleshooting content into the third language model, and the third language model summarizes the fault problem under the guidance of the third prompt word. The third prompt word is used to instruct the third language model to analyze and summarize the fault problem based on the troubleshooting results to draw a conclusion.

[0153] In an example, the third prompt word may be pre-set or obtained by training when training the third language model.

[0154] Exemplarily, the third prompt word may be:

[0155] The current stage is the summary stage. In the previous interactions, you have searched for certain information about the goals and problems given by the user through task decomposition + execution tools. You need to integrate this information and give the final conclusion in Chinese, and try to determine the root cause of the problem.

[0156] 1. The information being searched is obtained from many tools, which may lead to redundancy

[0157] 2. When the result is a question statement, directly output the result statement, do not create the content yourself

[0158] {memory}

[0159] Problem or goal: {goal}

[0160] Generate Chinese answers that are helpful to users. The answers should be concise and organized. You can use the markdown format and the font size should not be too large.

[0161] Here, {memory} can be a record of the processing result information of the above interaction stages (i.e., information extraction, troubleshooting process and sequence planning, and tool execution).

[0162] In the disclosed embodiment, a pre-trained fault location agent is used to summarize the troubleshooting results obtained through the trained third language model combined with the third prompt word, thereby improving the accuracy of the problem summary and achieving more accurate fault location.

[0163] For example, Figure 2 As shown in FIG, the implementation process of using the fault location agent to locate the fault is as follows: the user inputs the fault description information to the fault location agent. The fault location agent extracts the fault description information input by the user and obtains key information ( Figure 2 The key information is extracted from the fault location information) and the key information is recorded. Furthermore, the fault location agent determines the decision fault location chain, that is, determines the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information, and records the fault troubleshooting content and the execution order of each fault troubleshooting content. Then, the fault location agent calls the tool to execute, specifically calling the target detection tool corresponding to each fault troubleshooting content, performs fault troubleshooting based on the key information in the execution order, and obtains the troubleshooting results corresponding to each fault troubleshooting content (corresponding Figure 2 dotted box).

[0164] For the calling tool execution part, the target detection tools corresponding to each troubleshooting content can be called in the execution order, and troubleshooting is performed based on key information, that is, Figure 2 Run the fault location chain in . In this implementation process, the execution order is used to determine whether the tool chain has ended. When the tool chain ends, the troubleshooting results corresponding to each troubleshooting content are obtained, and the next step of summary analysis is entered. If the tool chain is not ended, the tool is called according to the execution order. During the tool call process, the target detection tool corresponding to the troubleshooting content is called. The target detection tool determines whether necessary parameters are missing, such as whether key information required for troubleshooting is missing. If the target detection tool does not lack necessary parameters, the troubleshooting task is executed (i.e. Figure 2 Run the tool in the test) and get the test result data ( Figure 2 The fault location agent context assignment in the middle, that is, the record of the detection result data), further, the fourth language model is called to analyze the detection result data to obtain the troubleshooting results corresponding to the troubleshooting content ( Figure 2 The tool output summary is in the tool output summary), and the troubleshooting results are recorded. If the target detection tool lacks required parameters, a context search is performed, that is, whether the context records the key information required for the execution of the target detection tool. If the query is found, the context search is successful, and the tool call is returned. If the query is not found, a vector representation of the key information extracted above is generated, and the similarity is calculated with the vector representation of the description information of the detection tool stored in the preset database. The vector representation of the description information of the detection tool stored in the preset database is used as the vector representation of the description information of the detection tool stored in the preset database. Figure 2 The detection tool corresponding to the vector library is determined as the final detection tool, and the tool call is returned; if the similarity is not greater than the set threshold ( Figure 2 The vector library matching fails). At this time, calling the tool to perform troubleshooting fails, and the next troubleshooting content is executed.

[0165] When the tool is called to execute and the troubleshooting results corresponding to each troubleshooting content are obtained, the fault location result is obtained according to each troubleshooting result ( Figure 2 (summarized and analyzed in).

[0166] The fault location agent outputs the obtained fault location results to the user.

[0167] In the disclosed embodiment, the fault location agent realizes fault location through four stages, providing a complete fault location mechanism. The fault location agent is implemented based on a large language model, and the fault location process is realized through natural language interaction, which simplifies the learning cost. For CDN scenarios, in CDN scenarios, because CDN is an edge node, there are many business and service change data, and the service request data is huge. If the manual fault location method is used, the troubleshooting efficiency is low. In comparison, the disclosed embodiment integrates the troubleshooting process of the operation and maintenance personnel into the function of the fault location agent, and locates the fault through the fault location agent, which improves the efficiency of fault location, thereby improving the stability of the business.

[0168] like Figure 3 As shown, the training process of the fault location agent in the embodiment of the present disclosure includes:

[0169] S301, obtaining sample fault description information, key annotation information corresponding to the sample fault description information, and a first sample prompt word.

[0170] The sample fault description information may be a device alarm or a fault information reported by a customer received by the operation and maintenance department. In one example, the sample fault description information may include, but is not limited to: the time when the abnormality occurred, the abnormality description information, the abnormality state, the location where the abnormality occurred, the abnormality domain name, etc. For the fault location agent, the sample fault description information is the user's query statement.

[0171] Key information is extracted or annotated from the sample fault description information to obtain annotated key information. The annotated key information may be, for example, information such as a domain name and region included in the sample fault description information.

[0172] The first sample prompt word is used to instruct the first large language model to extract key information from the sample fault description information.

[0173] For example, the sample fault description information may be: A customer domain name xxx.com reports a fault, and the connection quality in region B deteriorates. A first sample prompt word may be preset, and for example, the first sample prompt word may be the same as the first prompt word mentioned above.

[0174] S302: training a first language model based on the sample fault description information, the first sample prompt word, and the annotated key information.

[0175] The sample fault description information and the first sample prompt word are used as the input of the first language model. The output of the first language model is the predicted key information corresponding to the sample fault description information, and the labeled key information is used as the true value. By calculating the loss between the predicted key information and the labeled key information, the parameters of the first language model are adjusted to train the first language model. The trained first language model is used to extract information from the sample fault description information under the guidance of the first sample prompt word.

[0176] In one example, the first language model to be trained can be based on the pre-trained large language model and fine-tuned using sample fault description information, the first sample prompt words, and the annotated key information, so that the training speed of the first language model can be improved. During the training of the first language model, the first sample prompt words can also be adjusted to optimize the first sample prompt words.

[0177] S303, obtaining the annotated fault troubleshooting content corresponding to the sample fault description information, the annotated execution order of each annotated fault troubleshooting content, and the second sample prompt word.

[0178] For the sample fault description information, based on historical troubleshooting experience, set the second sample prompt word, the corresponding annotated troubleshooting content, and the annotated execution order of each annotated troubleshooting content. For example, the troubleshooting process in historical troubleshooting experience includes: determining the CDN node of the service, querying the service request log during the fault reporting period, querying whether the basic indicators of the service domain name are normal, querying whether there are problems with the quality of the CDN node of the service, and testing whether there are network problems with the CDN service node through ping, mtr, traceroute, etc. Correspondingly, the annotated troubleshooting content can be any one or more of the troubleshooting process.

[0179] Set the annotation execution order of each annotation troubleshooting content. The annotation execution order can be a step expression, which is used to indicate the sequence of the troubleshooting process. These troubleshooting processes can be executed in parallel, serially, or in combination of parallel and serial.

[0180] The second sample prompt word is used to instruct the second largest language model to plan the troubleshooting process and sequence for the sample fault description information, so as to determine the sample fault troubleshooting content corresponding to the sample fault description information and the execution order of the sample fault troubleshooting content, that is, to instruct the second largest language model to think about the troubleshooting content that needs to be troubleshooted for the sample fault description information and the execution order of these troubleshooting contents.

[0181] The second sample prompt word may be preset. Exemplarily, the second sample prompt word may be the same as the second prompt word mentioned above.

[0182] S304: training a second language model based on the sample fault description information, the second sample prompt word, the annotated fault troubleshooting content, and the annotation execution order of each annotated fault troubleshooting content.

[0183] The sample fault description information and the second sample prompt word are used as the input of the second largest language model. The output of the second largest language model is the predicted fault troubleshooting content corresponding to the sample fault description information and the predicted execution order of each predicted fault troubleshooting content. By calculating the loss between the predicted fault troubleshooting content and the labeled fault troubleshooting content, and between the predicted execution order of the predicted fault troubleshooting content and the labeled execution order of the labeled fault troubleshooting content, the parameters of the second largest language model are adjusted to train the second largest language model. The trained second largest language model is used to determine the fault troubleshooting content and its execution order corresponding to the sample fault description information under the guidance of the second sample prompt word.

[0184] In one example, the second largest language model to be trained can be based on the pre-trained large language model, and fine-tuned using sample fault description information, second sample prompt words, annotated fault troubleshooting content, and the order of annotation execution of each annotated fault troubleshooting content, so that the training speed of the second largest language model can be improved. In the process of training the second largest language model, the second sample prompt words can also be adjusted to optimize the second sample prompt words.

[0185] For example, the sample fault description information is: A customer domain name xxx.com reported a fault, and the quality of the connection in area B deteriorated. Under the instruction of the second sample prompt word, the output of the second largest language model can be expressed as:

[0186] - Thinking: First, I need to confirm whether there are any abnormalities in the basic information such as the overall bandwidth and number of requests of xxx.com business; at the same time, obtain the CDN node carrying information of the reporting domain name during the failure period, such as the changes in the scheduling coverage information of xxx.com's China Unicom in region B.

[0187] Step 1: Use the indicator detection tool to detect the bandwidth, number of requests, and status code indicator information of the domain name xxx.com in region B

[0188] Step 2: Use the scheduling detection tool to detect the current coverage changes of the domain name xxx.com in region B and the current bearer information

[0189] - Thinking: Now that I know the coverage and hosting service node information of the xxx.com domain name at the time of failure (current), I need to query the basic indicators of the CDN node, such as detection success rate, detection speed, node CPU pressure, etc., to see if there are any abnormalities; at the same time, obtain information on whether the node has abnormal changes at the same time; at the same time, obtain the service request log on the node and analyze whether there are any abnormalities

[0190] Step 3: Use the indicator detection tool to detect the current detection success rate, detection speed, and node CPU pressure indicator information of the xxx.com coverage node

[0191] Step 4: Use change control tools to detect recent abnormal changes in the xxx.com coverage node

[0192] Step 5: Use the log search tool to retrieve the log information of the domain name xxx.com at the above node

[0193] - Thinking: In order to achieve the goal, I can execute [Step 1] and [Step 2] at the same time. After these two steps are completed, I can execute [Step 3], [Step 4] and [Step 5]. So I can construct the operation logic like this:

[0194] Execution order: (Step1 AND Step2) -> (Step3 AND Step4 AND Step5).

[0195] Among them, the second sample prompt word may include information about tools that can be called, such as the above-mentioned {tool_specification}, so that the second largest language model can better think about the troubleshooting content that needs to be troubleshooted for the sample fault description information. The annotated troubleshooting content includes information about the detection tools required to be used for the annotated troubleshooting content. Exemplarily, the above-mentioned indicator detection tool, that is, the information about the detection tool required to be used for the troubleshooting content contained in the troubleshooting content Step1; the log retrieval tool, that is, the information about the detection tool required to be used for the troubleshooting content contained in the troubleshooting content Step5, and so on. The execution order indicates that the troubleshooting content Step1 is executed in parallel with Step2, Step3, Step4 and Step5 are executed in parallel, and (Step1 AND Step2) and (Step3 AND Step4 AND Step5) are executed in series.

[0196] In an example, the above steps S201 and S202 and steps S203 and S204 may be executed in parallel or sequentially.

[0197] S305 , calling the sample detection tools corresponding to each of the annotated troubleshooting contents respectively according to the annotated execution order, performing troubleshooting based on the annotated key information, and obtaining the sample troubleshooting results corresponding to each of the annotated troubleshooting contents.

[0198] The annotated troubleshooting content includes information about the detection tool required for the annotated troubleshooting content, and the sample detection tool corresponding to the annotated troubleshooting content can be determined through the detection tool information. Further, the sample detection tools corresponding to each annotated troubleshooting content are called in the order of annotation execution, and troubleshooting is performed based on the above annotated key information to obtain the sample troubleshooting results corresponding to each annotated troubleshooting content.

[0199] Exemplarily, the troubleshooting content is the above Step 1: Use the indicator detection tool to detect the bandwidth, number of requests, and status code indicator information of the domain name xxx.com currently connected to region B, and then call the indicator detection tool to perform troubleshooting on the basis of marking the key information domain name xxx.com and region B to obtain the corresponding sample troubleshooting results.

[0200] Among them, sample detection tools include the above-mentioned scheduling detection tools, indicator detection tools, log retrieval tools, etc.

[0201] The scheduling detection tool is used to detect the CDN node that actually provided the service at the time of the fault reporting customer, which is usually the first step in all fault reporting troubleshooting. Querying the CDN node that actually provides the service can provide a basis for subsequent analysis of node quality, node logs, etc. The scheduling detection tool will also check the scheduling switch before and after the fault occurs. If a scheduling switch occurs, it will also simultaneously analyze whether the switch caused the fault.

[0202] The indicator detection tool can detect all queries and analyses involving basic indicators. In actual applications, indicators usually include [bandwidth], [qps], [request success rate], [status code], etc.

[0203] The log search tool is used to search logs based on the time when the customer reported the problem and the customer's domain name.

[0204] S306, obtaining the labeled fault location result corresponding to each sample troubleshooting result and the third sample prompt word.

[0205] The third sample prompt word is used to instruct the third language model to analyze and summarize the fault problems of each sample troubleshooting result to draw a conclusion. The third sample prompt word can be preset, and illustratively, the third sample prompt word can be the same as the third prompt word mentioned above.

[0206] S307: training a third language model based on the sample inspection results, the third sample prompt words, and the annotated fault location results.

[0207] The sample inspection results and the third sample prompt words are used as the input of the third language model. The output of the third language model is the predicted fault location result corresponding to each sample inspection result. The labeled fault location result is used as the true value. The loss between the predicted fault location result and the labeled fault location result is calculated, and the parameters of the third language model are adjusted to train the third language model. The trained third language model is used to summarize the fault problems of each sample inspection result under the guidance of the third sample prompt words.

[0208] In one example, the trained third language model can be based on the pre-trained large language model and fine-tuned using the sample troubleshooting results, the third sample prompt words, and the labeled fault location results, so that the training speed of the third language model can be improved. During the training of the third language model, the third sample prompt words can also be adjusted to optimize the third sample prompt words.

[0209] S308, the trained fault location agent includes a first language model, a second language model and a third language model.

[0210] By integrating the first language model, the second language model and the third language model trained above with the calling and executing process of the detection tool, the function of the fault location agent is formed, and a trained fault location agent is obtained. Accordingly, the trained fault location agent includes the first language model, the second language model and the third language model trained above, and also includes the calling and executing of the detection tool.

[0211] In the disclosed embodiment, by training the first language model, the second language model and the third language model, the troubleshooting process of the operation and maintenance personnel is integrated into the function of the fault location intelligent agent, so that the fault can be located through the fault location intelligent agent, thereby improving the efficiency of fault location.

[0212] The training method of the fault location agent in the embodiment of the present disclosure is as follows: Figure 4 As shown, the fault location agent includes a first language model, a second language model and a third language model, and the training method includes:

[0213] S401, obtaining sample fault description information, key information marked corresponding to the sample fault description information, and a first sample prompt word.

[0214] S402: training a first language model based on the sample fault description information, the first sample prompt word, and the annotated key information.

[0215] The first language model is used to extract information from sample fault description information under the guidance of the first sample prompt word.

[0216] S403, obtaining the annotated fault troubleshooting content corresponding to the sample fault description information, the annotated execution order of each annotated fault troubleshooting content, and the second sample prompt word.

[0217] S404: training a second language model based on the sample fault description information, the second sample prompt word, the annotated fault troubleshooting content, and the annotation execution order of each annotated fault troubleshooting content.

[0218] The second largest language model is used to determine the troubleshooting content and its execution order corresponding to the sample fault description information under the guidance of the second sample prompt word; the annotated troubleshooting content includes information about the detection tool required to use the annotated troubleshooting content.

[0219] In an example, the above steps S401 and S402 and steps S403 and S404 may be executed in parallel or sequentially.

[0220] S405 , calling the sample detection tools corresponding to each of the annotated troubleshooting contents respectively according to the annotated execution order, performing troubleshooting based on the annotated key information, and obtaining the sample troubleshooting results corresponding to each of the annotated troubleshooting contents.

[0221] S406, obtaining the labeled fault location result corresponding to each sample troubleshooting result and the third sample prompt word.

[0222] S407: training a third language model based on the sample inspection results, the third sample prompt words, and the annotated fault location results.

[0223] The third language model is used to summarize the fault problems of each sample troubleshooting result under the guidance of the third sample prompt word.

[0224] In the disclosed embodiment, by training the first language model, the second language model and the third language model, the troubleshooting process of the operation and maintenance personnel is integrated into the function of the fault location intelligent agent, so that the fault can be located through the fault location intelligent agent, thereby improving the efficiency of fault location.

[0225] The present disclosure also provides a fault location device based on a large model, see Figure 5 ,include:

[0226] Information acquisition module 501, used to acquire fault description information;

[0227] The information extraction module 502 is used to extract the fault description information using the pre-trained fault location agent to obtain key information;

[0228] The content determination module 503 is used to determine the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information by using the pre-trained fault location agent;

[0229] The first troubleshooting module 504 is used to use the pre-trained fault location agent to call the target detection tool corresponding to each troubleshooting content, perform troubleshooting based on key information in the execution order, and obtain the troubleshooting results corresponding to each troubleshooting content;

[0230] The fault location module 505 is used to obtain a fault location result based on each troubleshooting result by using a pre-trained fault location agent.

[0231] In a possible implementation, the information extraction module 502 is specifically used to utilize a pre-trained fault location agent, call the first language model, and input the fault description information into the first language model to extract information under the guidance of the first prompt word to obtain key information.

[0232] In a possible implementation, the above-mentioned content determination module 503 is specifically used to utilize a pre-trained fault location agent, call the second largest language model, and input the fault description information into the second largest language model to determine the fault troubleshooting content and the execution order of the fault troubleshooting content under the guidance of the second prompt word.

[0233] In a possible implementation, the above troubleshooting contents have levels and priorities, and troubleshooting contents at the same level are executed in parallel, while troubleshooting contents at different priorities are executed serially.

[0234] In a possible implementation manner, the above-mentioned troubleshooting content includes information about a detection tool required to be used for the troubleshooting content; the above-mentioned first troubleshooting module 504 includes:

[0235] The tool determination submodule is used to use the pre-trained fault location agent to determine the target detection tool corresponding to the fault troubleshooting content according to the information of the detection tool contained in the fault troubleshooting content;

[0236] The troubleshooting submodule is used to call the target detection tools corresponding to each troubleshooting content in the execution order, perform troubleshooting based on key information, and obtain troubleshooting results corresponding to each troubleshooting content.

[0237] In a possible implementation, the above-mentioned troubleshooting sub-module is specifically used to call the target detection tool corresponding to the troubleshooting content, so that the target detection tool performs troubleshooting based on key information, obtains detection result data, and calls the fourth language model to analyze the detection result data to obtain the troubleshooting results corresponding to the troubleshooting content.

[0238] In a possible implementation, the above-mentioned fault location module 505 is specifically used to utilize a pre-trained fault location agent to call the third largest language model, and input the troubleshooting results corresponding to each fault troubleshooting content into the third largest language model to summarize the fault problem under the guidance of the third prompt word to obtain the fault location result.

[0239] In a possible implementation manner, the above device further includes:

[0240] A fourth acquisition module, used to acquire sample fault description information, annotated key information corresponding to the sample fault description information, and a first sample prompt word;

[0241] The fourth training module is used to train the first language model based on the sample fault description information, the first sample prompt word and the marked key information; the first language model is used to extract information from the sample fault description information under the guidance of the first sample prompt word;

[0242] A fifth acquisition module, used to acquire the annotated fault troubleshooting content corresponding to the sample fault description information and the annotated execution order of each annotated fault troubleshooting content, and the second sample prompt word;

[0243] The fifth training module is used to train the second largest language model based on the sample fault description information, the second sample prompt word, the labeled fault troubleshooting content and the labeled execution order of each labeled fault troubleshooting content; the second largest language model is used to determine the fault troubleshooting content corresponding to the sample fault description information and its execution order under the guidance of the second sample prompt word; the labeled fault troubleshooting content includes information on the detection tool required to be used for the labeled fault troubleshooting content;

[0244] The third troubleshooting module is used to call the sample detection tools corresponding to each marked troubleshooting content according to the marked execution order, perform troubleshooting based on the marked key information, and obtain the sample troubleshooting results corresponding to each marked troubleshooting content;

[0245] A sixth acquisition module, used to acquire the labeled fault location result corresponding to each sample troubleshooting result and the third sample prompt word;

[0246] The sixth training module is used to train the third language model based on the sample troubleshooting results, the third sample prompt words and the labeled fault location results; the third language model is used to summarize the fault problems in the sample troubleshooting results under the guidance of the third sample prompt words; the trained fault location intelligent agent includes the first language model, the second language model and the third language model.

[0247] The present disclosure also provides a training device for a fault location agent. Figure 6The fault location agent includes a first language model, a second language model and a third language model, and the training device includes:

[0248] A first acquisition module 601 is used to acquire sample fault description information, key information marked corresponding to the sample fault description information, and a first sample prompt word;

[0249] The first training module 602 is used to train a first language model based on the sample fault description information, the first sample prompt word and the annotation key information; the first language model is used to extract information from the sample fault description information under the guidance of the first sample prompt word;

[0250] The second acquisition module 603 is used to acquire the annotated fault troubleshooting content corresponding to the sample fault description information and the annotated execution order of each annotated fault troubleshooting content, as well as the second sample prompt word;

[0251] The second training module 604 is used to train the second largest language model based on the sample fault description information, the second sample prompt word, the labeled fault troubleshooting content and the labeled execution order of each labeled fault troubleshooting content; the second largest language model is used to determine the fault troubleshooting content corresponding to the sample fault description information and its execution order under the guidance of the second sample prompt word; the labeled fault troubleshooting content includes information about the detection tool required to be used for the labeled fault troubleshooting content;

[0252] The second troubleshooting module 605 is used to call the sample detection tools corresponding to each of the marked troubleshooting contents according to the marked execution order, perform troubleshooting based on the marked key information, and obtain the sample troubleshooting results corresponding to each of the marked troubleshooting contents;

[0253] The third acquisition module 606 is used to obtain the marked fault location result corresponding to each sample troubleshooting result and the third sample prompt word;

[0254] The third training module 607 is used to train the third language model based on the sample troubleshooting results, the third sample prompt words and the annotated fault location results; the third language model is used to summarize the fault problems of the sample troubleshooting results under the guidance of the third sample prompt words.

[0255] The present disclosure also provides an intelligent agent, see Figure 7 ,include:

[0256] Input module 701, used to receive fault description information;

[0257] The processing module 702 is used to extract information from the fault description information received by the input module to obtain key information; determine the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information; call the target detection tool corresponding to each fault troubleshooting content, perform fault troubleshooting based on the key information in the execution order, and obtain the troubleshooting results corresponding to each fault troubleshooting content; obtain the fault location result based on each troubleshooting result;

[0258] The output module 703 is used to output the fault location result obtained by the processing module.

[0259] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0260] The present disclosure provides an electronic device, comprising:

[0261] at least one processor; and

[0262] a memory communicatively connected to at least one processor; wherein,

[0263] The memory stores instructions that can be executed by at least one processor. The instructions are executed by at least one processor to enable the at least one processor to perform any method in the present disclosure.

[0264] A non-transitory computer-readable storage medium storing computer instructions is disclosed, wherein the computer instructions are used to enable a computer to execute any method in the present disclosure.

[0265] The present disclosure provides a computer program product, including a computer program, which implements any method in the present disclosure when executed by a processor.

[0266] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0267] Figure 8 A schematic block diagram of an example electronic device 800 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0268] like Figure 8 As shown, the device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 to a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0269] A number of components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0270] The computing unit 801 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 801 performs the various methods and processes described above, such as a fault location method based on a large model and a training method for a fault location agent. For example, in some embodiments, a fault location method based on a large model and a training method for a fault location agent may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the fault location method based on a large model and the training method for a fault location agent described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured in any other appropriate manner (for example, by means of firmware) to execute the large model-based fault localization method and the fault localization agent training method.

[0271] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0272] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0273] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0274] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0275] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0276] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0277] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0278] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A fault location method based on a large model, comprising: Get fault description information; Using a pre-trained fault location agent, extract information from the fault description information to obtain key information; Using a pre-trained fault location agent, determine the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information; Using the pre-trained fault location agent, calling the target detection tool corresponding to each of the troubleshooting contents, performing troubleshooting based on the key information in the execution order, and obtaining troubleshooting results corresponding to each of the troubleshooting contents; A pre-trained fault location agent is used to obtain a fault location result based on each of the troubleshooting results.

2. The method according to claim 1, wherein: The pre-trained fault location agent is used to extract information from the fault description information to obtain key information, including: The pre-trained fault location agent is used to call the first language model, and the fault description information is input into the first language model to extract information under the guidance of the first prompt word to obtain key information.

3. The method according to claim 1, wherein: The method of using a pre-trained fault location agent to determine the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information includes: The pre-trained fault location agent is used to call the second largest language model, and the fault description information is input into the second largest language model to determine the fault troubleshooting content and the execution order of the fault troubleshooting content under the guidance of the second prompt word.

4. The method according to claim 3, wherein: The troubleshooting contents have levels and priorities. Troubleshooting contents at the same level are executed in parallel, and troubleshooting contents at different priorities are executed serially.

5. The method according to claim 1, wherein: The troubleshooting content includes information about the detection tools required for the troubleshooting content; The pre-trained fault location agent is used to call the target detection tool corresponding to each of the troubleshooting contents, and troubleshooting is performed based on the key information in the execution order to obtain the troubleshooting results corresponding to each of the troubleshooting contents, including: Using a pre-trained fault location agent, according to the information of the detection tool contained in the fault troubleshooting content, determine the target detection tool corresponding to the fault troubleshooting content; The target detection tools corresponding to each of the troubleshooting contents are respectively called in the execution order, troubleshooting is performed based on the key information, and troubleshooting results corresponding to each of the troubleshooting contents are obtained.

6. The method according to claim 5, wherein: Calling the target detection tool corresponding to the troubleshooting content, performing troubleshooting based on the key information, and obtaining troubleshooting results corresponding to the troubleshooting content, including: Calling a target detection tool corresponding to the troubleshooting content, so that the target detection tool performs troubleshooting based on the key information to obtain detection result data, and calling a fourth language model to analyze the detection result data to obtain a troubleshooting result corresponding to the troubleshooting content.

7. The method according to claim 1, wherein: The method of using the pre-trained fault location agent to obtain a fault location result based on each of the troubleshooting results includes: The pre-trained fault location agent is used to call the third language model, and the troubleshooting results corresponding to each of the troubleshooting contents are input into the third language model to summarize the fault problems under the guidance of the third prompt word to obtain the fault location result.

8. The method according to any one of claims 1 to 7, wherein: The training process of the fault location agent includes: Acquire sample fault description information, annotated key information corresponding to the sample fault description information, and a first sample prompt word; Based on the sample fault description information, the first sample prompt word and the labeled key information, a first large language model is trained; the first large language model is used to extract information from the sample fault description information under the guidance of the first sample prompt word; Obtaining the annotated fault troubleshooting content corresponding to the sample fault description information and the annotated execution order of each annotated fault troubleshooting content, as well as a second sample prompt word; Based on the sample fault description information, the second sample prompt word, the annotated fault troubleshooting content and the annotated execution order of each annotated fault troubleshooting content, a second language model is trained; the second language model is used to determine the fault troubleshooting content corresponding to the sample fault description information and its execution order under the guidance of the second sample prompt word; the annotated fault troubleshooting content includes information about the detection tool required to be used for the annotated fault troubleshooting content; Invoke the sample detection tools corresponding to the troubleshooting contents of the annotations respectively according to the annotation execution order, perform troubleshooting based on the annotation key information, and obtain the sample troubleshooting results corresponding to the troubleshooting contents of the annotations; Obtaining the labeled fault location result corresponding to each of the sample troubleshooting results and the third sample prompt word; Based on the sample troubleshooting results, the third sample prompt word and the annotated fault location result, a third language model is trained; the third language model is used to summarize the fault problems of the sample troubleshooting results under the guidance of the third sample prompt word; The trained fault location agent includes the first largest language model, the second largest language model and the third largest language model.

9. A method for training a fault localization agent, the fault localization agent comprising a first language model, a second language model and a third language model, the method comprising: Acquire sample fault description information, annotated key information corresponding to the sample fault description information, and a first sample prompt word; Based on the sample fault description information, the first sample prompt word and the labeled key information, a first large language model is trained; the first large language model is used to extract information from the sample fault description information under the guidance of the first sample prompt word; Obtaining the annotated fault troubleshooting content corresponding to the sample fault description information and the annotated execution order of each annotated fault troubleshooting content, as well as a second sample prompt word; Based on the sample fault description information, the second sample prompt word, the annotated fault troubleshooting content and the annotated execution order of each annotated fault troubleshooting content, training a second language model; the second language model is used to determine the fault troubleshooting content corresponding to the sample fault description information and its execution order under the guidance of the second sample prompt word; The annotated troubleshooting content includes information about the detection tool required to be used for the annotated troubleshooting content; Invoke the sample detection tools corresponding to the troubleshooting contents of the annotations respectively according to the annotation execution order, perform troubleshooting based on the annotation key information, and obtain the sample troubleshooting results corresponding to the troubleshooting contents of the annotations; Obtaining the labeled fault location result corresponding to each of the sample troubleshooting results and the third sample prompt word; Based on the sample troubleshooting results, the third sample prompt words and the labeled fault location results, a third language model is trained; the third language model is used to summarize the fault problems in the sample troubleshooting results under the guidance of the third sample prompt words.

10. A fault location device based on a large model, comprising: An information acquisition module is used to obtain fault description information; An information extraction module, used to extract information from the fault description information using a pre-trained fault location agent to obtain key information; A content determination module, used to determine the fault troubleshooting content and the execution order of each fault troubleshooting content based on the fault description information by using a pre-trained fault location agent; A first troubleshooting module is used to use a pre-trained fault location intelligent agent to call a target detection tool corresponding to each of the troubleshooting contents, perform troubleshooting based on the key information in the execution order, and obtain troubleshooting results corresponding to each of the troubleshooting contents; The fault location module is used to obtain a fault location result based on each of the above-mentioned troubleshooting results by using a pre-trained fault location agent.

11. A training device for a fault location agent, the fault location agent comprising a first large language model, a second large language model and a third large language model, the training device comprising: A first acquisition module, used to acquire sample fault description information, annotated key information corresponding to the sample fault description information, and a first sample prompt word; A first training module, used for training a first large language model based on the sample fault description information, the first sample prompt word and the annotation key information; the first large language model is used for extracting information from the sample fault description information under the guidance of the first sample prompt word; A second acquisition module is used to acquire the annotated fault troubleshooting content corresponding to the sample fault description information and the annotated execution order of each annotated fault troubleshooting content, as well as a second sample prompt word; A second training module is used to train a second large language model based on the sample fault description information, the second sample prompt word, the labeled fault troubleshooting content and the labeled execution order of each labeled fault troubleshooting content; the second large language model is used to determine the fault troubleshooting content corresponding to the sample fault description information and its execution order under the guidance of the second sample prompt word; The annotated troubleshooting content includes information about the detection tool required to be used for the annotated troubleshooting content; A second troubleshooting module is used to call the sample detection tools corresponding to each of the annotated troubleshooting contents respectively according to the annotated execution order, perform troubleshooting based on the annotated key information, and obtain sample troubleshooting results corresponding to each of the annotated troubleshooting contents; A third acquisition module is used to acquire the marked fault location result corresponding to each of the sample troubleshooting results and the third sample prompt word; The third training module is used to train a third language model based on the sample troubleshooting results, the third sample prompt words and the labeled fault location results; the third language model is used to summarize the fault problems of each sample troubleshooting result under the guidance of the third sample prompt words.

12. An intelligent agent, comprising: An input module, used for receiving fault description information; A processing module, used for extracting the fault description information received by the input module to obtain key information; Determine the troubleshooting content and the execution order of each troubleshooting content based on the fault description information; call the target detection tool corresponding to each troubleshooting content, perform troubleshooting based on the key information in the execution order, and obtain the troubleshooting results corresponding to each troubleshooting content; Obtaining a fault location result based on each of the troubleshooting results; An output module is used to output the fault location result obtained by the processing module.

13. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 9.

14. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-9.

15. A computer program product, comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 9.

Images